AI as a tool vs AI as a product

Ever since discourse around AI has come to prominence, I’ve always believed this one thing: AI is a great tool, but AI is a lousy product.

Example: Today, I had to recall exactly when my employer changed its 401k provider and which company we changed from. Because we’re a private company, there is no “public” record of these things, but all communication related to this event is buried somewhere in my email archives. First, I had to figure out exactly when this event occurred. I knew it was before COVID, but was it 2018 or 2017? I couldn’t remember. A simple search for 401k brought up a few office email threads, but also every marketing or account statement email I’ve ever received on this topic. OK, well then… narrow it down. Email from our office manager, containing keywords 401k. That got me a few emails that were relevant, but clearly after the change had happened and didn’t include the name of the service provider we had been using previously (which is what I really needed to know). Hm… so maybe the email that contains the relevant information wasn’t originally from my work but from the provider? Search for that, get nothing relevant. But also, I see the term “retirement account” used a lot. OK, maybe from our office manager, containing the word “retirement” from before 2020. That produced a few results, containing many PDF attachments that I had to open and read through, but eventually came across the answer I was looking for: Date the change happened, from Company X to Company Y. Boom.

This is all the digital equivalent of sifting through filing cabinets of paper folders that would contain the same information… but it shouldn’t be. A suitably advanced LLM running on my local machine that had access to my email archives and messaging history could have found this answer for me likely within seconds.

We don’t want artificially generated cartoon graphics. We want Jarvis for our email!